Senior & Lead Software Engineer and Architect
I build software that has to run correctly and invisibly: measurement platforms and cross-browser extensions used by millions at Comscore and Nielsen, low-latency trading infrastructure at Morgan Stanley. Today I bring that systems discipline to production AI applications on AWS Bedrock.
Twenty-plus years of building systems that have to be right the first time.
I build software that has to run correctly and invisibly — on millions of machines, inside strict resource budgets, under real privacy constraints. Over twenty years that has meant audience measurement, low-latency trading, and, most recently, applied AI.
Now. Since mid-2025 I have worked independently, shipping production AI applications on Amazon Bedrock: SecureView, a privacy-first Chrome extension that categorises browsing without ever sending the same domain twice; SecureBin, an Australian waste-classification app built on Bedrock vision; and two more — each on a real serverless pipeline, with keys server-side and data retention bounded by design.
Before that. I engineered audience measurement for Comscore, and for Nielsen via TCS — Windows applications and cross-browser extensions used by millions, including the award-winning Netsight Chromebook Meter, which I designed from concept through to production. Earlier, I built exchange-connectivity infrastructure for Morgan Stanley's equity trading systems, and co-developed Vaachak, a Hindi and English text-to-speech engine that won a National Award for the Empowerment of Persons with Disabilities, built at Prologix — later acquired by Nuance, now part of Microsoft.
What I am looking for. Senior engineering and architect roles, based in Melbourne, where deep systems experience matters as much as modern AI capability. Reliability, performance budgets, and the awkward edge cases are the parts of the job I enjoy most.
What I reach for, from systems-level C++ through to serverless AI on AWS.
Formal credentials behind the applied-AI work.
Newest first — independent AI work today, back through audience measurement, equity trading, and speech research.
Designing and shipping production-grade AI applications on AWS Bedrock full time, deepening hands-on applied-AI capability, since my role at Comscore was made redundant in June 2025.
A Chrome Extension (Manifest V3) that tracks active browsing time per site, automatically categorises every domain into one of 11 categories (Technology, Entertainment, Productivity, and more), and surfaces the data through a clean popup with daily history and search. For sites that cannot be classified by rule, it falls back to Amazon Bedrock via a serverless AWS pipeline — keeping the API key entirely server-side. Built with pure vanilla JavaScript; no build step, no dependencies.
Chrome Extension rule engine (local, no network) → cache miss triggers AI path: CloudFront → Lambda@Edge (auth) → API Gateway → Lambda → Amazon Bedrock (Claude) → category returned → cached in Chrome storage so the same domain is never sent twice.
A PyTorch image-classification app that identifies flower species from a photograph and returns top-K predictions with confidence scores, trained on the 102 Category Flower Dataset using transfer learning over a pre-trained backbone with a custom classification head. Features dynamic dataset sizing, automated best-model checkpointing, and a FastAPI backend packaged as a container image and served from a Lambda Function URL — no servers to keep warm between requests. Completed as part of RMIT's AI Programming with Python Nanodegree.
Browser UI → Lambda Function URL (HTTPS) → AWS Lambda running a container image pulled from ECR (FastAPI via Mangum) → PyTorch inference → top-K species + confidence returned.
A cross-platform iOS & Android app (React Native / Expo) that uses the device camera to photograph waste
items and classifies which bin they belong to using Amazon Bedrock's vision AI.
Advice is tailored to the user's Australian state and council — covering all eight kerbside bin categories
(general waste, mixed recycling, organics/FOGO, glass, drop-off, reuse, and more). Scan history is stored
locally on-device. No login or account required. Images are deleted from S3 immediately after processing;
job records in DynamoDB expire after 24 hours.
*Disclaimer: SecureBin is an independent application and is not affiliated with any government entity.
Mobile app → CloudFront + OAC (SigV4) → Lambda presign URL → direct S3 PUT → S3 event triggers Processor Lambda → Amazon Bedrock APAC cross-region inference (Sydney + Melbourne) → DynamoDB result → app polls → displays item name + bin colour.
A practice-test app for primary school kids, pitched at the Australian ICAS and NAPLAN assessments. A child picks a year level and a subject, works through up to 30 questions spread evenly across the subject's topics, and gets a result with an explanation for every answer. Review mode reveals each explanation as the question is answered; Exam mode holds everything back until submission — and that rule is enforced server-side, so the API simply never sends the correct answers until the play is complete. There are no API keys or bearer tokens in the browser: Cognito issues short-lived AWS credentials, every request is SigV4-signed, and each Lambda scopes its DynamoDB queries to the caller identity that API Gateway supplies rather than to any client-provided user id.
React + Vite SPA (S3/CloudFront) → Cognito User Pool → Identity Pool issues temporary AWS credentials → SigV4-signed requests → API Gateway (AWS_IAM authorizer) → Lambda (Python 3.11) → DynamoDB (question bank + per-user test history, with Point-in-Time Recovery).
A global media measurement and analytics company that provides marketing data and analytics to enterprises, advertising agencies, brand marketers, and publishers. www.comscore.com
ResearchSecure is a Windows-based tracking application deployed on panelist machines to silently monitor and capture user activity for behavioural research and analytics. I was responsible for designing, implementing, maintaining, and enhancing critical application modules to enable secure, efficient, and unobtrusive data collection. My work involved close collaboration with cross-functional teams to improve the application's performance, reliability, and scalability, all while ensuring strict adherence to data privacy and security standards.
Additionally, contributed to the maintenance and support of Comscore's legacy Windows applications, ensuring stability, compatibility, and ongoing performance improvements.
TCS: Part of the Tata group and one of the world's largest IT services and consulting firms. www.tcs.com
Nielsen: A global leader in online market research, statistical measurement and analytics. www.nielsen.com
Netsight Chromebook Meter is a key component of Nielsen's Audience Measurement suite within the Nielsen Digital Panel portfolio. It is designed to unobtrusively capture detailed panelist activity on Chromebooks without disrupting the user experience. The collected behavioural data is securely transmitted to AWS S3 for further processing and storage.
Netsight Meter is a core component of Nielsen's Audience Measurement platform under the Nielsen Digital Panel portfolio. This Windows-based application is designed to unobtrusively log detailed panelist activity on their computers without disrupting their experience. Collected data is securely uploaded to AWS S3 for processing and storage.
TCS (Tata group) for Nielsen — the same Digital Panel programme as the Australian engagement above, six years earlier.
Netsight Meter is a core component of Nielsen's Audience Measurement platform under the Nielsen Digital Panel portfolio. This Windows-based application is designed to unobtrusively log detailed panelist activity on their computers without disrupting their experience.
TCS: Part of the Tata group, one of the world's largest IT services firms. www.tcs.com
Morgan Stanley: A global investment firm specialising in wealth management, investment banking, and sales and trading services. www.morganstanley.com
Exchange Connectivity (Order Router-OR) connects the Morgan Stanley internal systems to Trading Exchanges across Asian, European and US regions.
A product-based company working in the area of Speech Synthesis, later acquired by Nuance/Microsoft (a leading global company that provides speech and imaging applications). www.nuance.com
Vaachak is a Windows/Linux based text-to-speech (TTS) engine developed using C++. TTS is a speech conversion technology that enables automatic systems to convert text to natural-sounding speech. Prologix was first acquired by Nuance, which was later acquired by Microsoft.
A Gov. of India research company working in the area of Speech Recognition, later merged with its Pilani Centre. www.ceeri.res.in
Spoon Feeding Robot System was a speech-enabled CSIR Network project, developed to assist physically challenged persons with feeding.
Peer-reviewed work on speech recognition and synthesis, published 2004–2010.
Formal study, then and now.
Recognition from Nielsen, Morgan Stanley, TCS, and the Government of India.
Open to senior engineering and architect roles — Melbourne or remote.